Re: [Libunwind-devel] the behavior of unw_set_caching_policy

[Masatake YAMATO]

Thank you for replying.

You wrote "if you're unwinding from the same IP address repeatedly."    

About strace my understanding is that invalidation is required when
the target process calls system call which modifies the address space
like mmap and execve because even IP address is the same IP can point 
different library or executable. Is this correct?

Masatake YAMATO

On Wed, 11 Dec 2013 08:33:31 +0530, Arun Sharma <address@hidden> wrote:
> On Tue, Dec 10, 2013 at 5:06 PM, Masatake YAMATO <address@hidden> wrote:
>> What kind of things are cached with the interface unw_set_caching_policy?
>>
>> As I wrote ago I'm working on adding stack trace feature to strace.
>> For implementing the feature libunwind-ptrace is used.
>> Should I turing on the cache mechanism with unw_set_caching_policy
>> in the case? As far as I trying, the cache improves the performance.
>> However, I wonder what I should invalidate the cache.
> 
> Mainly the dwarf_rs_cache. Helps avoid reparsing the dwarf unwind
> info, if you're unwinding from the same IP address repeatedly.
> If you generate JIT'ed code and the unwind info associated with the
> code changes, that's when you're required to invalidate.
> 
>  -Arun